Could Male Fertility be Predicted with a Simple Questionnaire?

Photo by Kelly Sikkema on Unsplash

Infertility in men is an issue seldom spoken of in society. Although we know there are many factors involved in men’s lives that can affect how fertile they are, we can only know for sure through analysis. However, I would like to pose the questions: What if a simple questionnaire were enough? What if by filling in a form of several questions about one’s lifestyle and medical history, and that would be enough to accurately gauge male fertility?

Well, by analysing a set of 100 males, who each answered a questionnaire, had their semen analysed according to the WHO 2010 criteria and were diagnosed as either ‘Normal quality’ or ‘Altered quality’. We could, potentially, build a model to predict whether a man has ‘Normal’ or ‘Altered’ semen using the answers provided by the questionnaire. But, first let us view some information about the men in the study:

(The Questionnaire asked how many hours a day do you sit down, and the men answered with the number of hours so, I plotted a density curve to see how many hours the group generally sat down. And the graph on the right show each age group and the number of men in that age group.)

Above on the left I have plotted a density curve to show the length of time that most men sat down per day on average for, and the ages of the men who took part in the study. We can now compare the diagnoses that men received with their answers to the questions. It is also worth noting that only men aged 27 to 36 were included in this set.

(3 three bar charts: First plots number of men and their alcohol consumption (with 0.2 being several times a day and 1 being never/hardly ever). Second shows smoking habits (-1 being never 1 being several times a day). And the third shows whether a man had a Childhood disease (i.e., chicken pox, measles, mumps, polio) (1 being yes and 0 being no). Each graph split the count between ’N’ for Normal diagnosis and ‘O’ for altered diagnosis.)

Due to the small size of the dataset, we cannot say much about the alcohol consumption chart as there is only one person who admitted to drinking several times a day and several times a week. But when comparing the proportion of men who received a diagnosis that stated there were no alterations with the men who did have alterations: as alcohol consumption increased, this ratio fell (excluding the first two groups), implying that alcohol is potentially harmful to fertility. The same can be said for smoking frequency, and whether a man had childhood disease or not. I have omitted plots and explanations on the other variables because they did not seem to affect diagnoses significantly. That said, they were included when training the model to predict a diagnosis.

I used a KNN model with k = 3 and the accuracy, precision and recall were 0.83, 0.96 and 0.857, respectively. The model was pretty good at predicting the correct diagnosis. However, the dataset is extremely small so this model is unlikely to perform this well for a larger group. Also, the dataset is not balanced, as there were only 12 men who received an ‘Altered Diagnosis,’ so the model had very little to learn from/compare to. I also mentioned earlier that only men aged 27 to 36 were in this particular dataset meaning it wouldn’t be very representative of all men, nor would the model be able to accurately gauge effects of age ion fertility. It still may be possible to get a model that is more accurate and robust, given a larger dataset. Such an advancement would mean a lot less time & resources fertility clinics would need to give men a diagnosis, and less time for men to anxiously await results. So, it is worth researching with a larger group of volunteers.

[1]David Gil, Jose Luis Girela, Joaquin De Juan, M. Jose Gomez-Torres, and
Magnus Johnsson. Predicting seminal quality with artificial intelligence
methods. Expert Systems with Applications, 39(16):12564 – 12573, 2012. Available at:

I am a recent Mathematics graduate with interest in developing in the fields of Machine Learning and Data Science.